DyeDactic: towards biosynthetic alternatives to artificial textile dyes

Production and application of textile dyes using microorganisms represents an important step towards sustainable manufacturing. Although living organisms can produce numerous coloured substances, they frequently demonstrate poor dyeing performance, insufficient photostability, or toxicity. To guide the development of new biosynthetically accessible colourants, we developed a workflow (DyeDactic) to predict colour at different pH values. Starting with a rapid estimation of the lowest electronic transition energy for potential colourants using Graph Neural Networks, the procedure can filter large libraries of generated chemical structures producing a targeted subset of compounds for further examination. The final step employs time-dependent density functional theory (TD-DFT) to estimate the intensity of absorption peaks in the visible spectral region, model spectral band shape and estimate colour. To tackle halochromism, which is frequently observed for natural colourants, populations of protonated and deprotonated species are estimated at different pH values using predicted acidity constants of ionisable atoms followed by addition and weighting of modelled absorption spectra. The complete workflow is applied to four natural colourants belonging to different classes (emodin, quinalizarin, biliverdin, and orcein) and the predicted colour dependence on pH is compared with the experimental data. Both the machine learning tool and the quantum chemical calculations are validated and fine-tuned using an assembled dataset of spectral properties of 647 natural colourants. Potential chemoenzymatic modifications are discussed based on comparison of structural and physico-chemical properties between natural colourants and artificial dyes and pigments from the Colour Index.